Karpathy-style graph engineering turns multi-agent loops into shared-memory systems
leslysandra · x · 2026-07-27
Karpathy-style graph engineering proposes shared memory for multi-agent loops
The post highlights a 12-page “Graph Engineering” playbook for multi-agent systems. The core idea is to move from a single generate/critic/revise loop to parallel agents working in separate worktrees, then add a knowledge graph that stores entities, aliases, typed edges, and facts.
Key claims in the workflow:
- The graph becomes shared memory that workers can write to and evaluators can query.
- Evaluators should check claims against graph edges rather than “vibes.”
- This lets loops persist overnight instead of rebuilding context every session.
- The framing is that the agent forgets, but the graph does not.
More from coding & agent
- Agentic Data Science in Practice: Agents Write Code but Answer Wrong Questions — hugobowne · 2026-07-27
- A VS Code extension adds Markdown-style highlighting to Alchemy string templates — samgoodwin89 · 2026-07-27
- Claude’s Stripe MCP connector is being called unusable after repeated disconnects — evielync · 2026-07-27
- Reddit user says Vecel’s Eve looks like “Next.js for agents” — richie9830 · 2026-07-27
- Leader.skill open-sources a framework for turning vague requests into executable agent goals — 数字生命卡兹克 · 2026-07-27
- MiniBot 2.40 adds xAI, HF Studio and vLLM support with inline media tools — Creative-Type9411 · 2026-07-27